Variations in snow strength and stability on uniform slopes · Variations in snow strength and stability on uniform slopes C. Landrya,*, K. Birkelanda,b, K. Hansena, ... Seven slopes

www.elsevier.com/locate/coldregions

Cold Regions Science and Technology 39 (2004) 205–218

Variations in snow strength and stability on uniform slopes

C. Landrya,*, K. Birkelanda,b, K. Hansena, J. Borkowskic, R. Brownd, R. Aspinalla

aDepartment of Earth Sciences, Montana State University, P.O. Box 366, Silverton, CO 81433, USAbU.S. Forest Service National Avalanche Center, USA

cDepartment of Mathematical Sciences, Montana State University, USAdDepartment of Civil Engineering, Montana State University, USA

Received 30 September 2002; accepted 30 November 2003

Abstract

This research investigated whether single snowpits could reliably represent snowpack strength and stability conditions

throughout apparently ‘uniform’ slopes. Seven slopes were selected by experienced avalanche forecasters, three each in the

Bridger and Madison Ranges of Southwest Montana (USA), and one in the Columbia Mountains near Rogers Pass, British

Columbia (Canada). Teams performed 10 quantified loaded column tests in each of five snowpits within a 900 m2 sampling site

on each ‘uniform’ slope, measuring shear strength in a single weak layer. Collection of slab shear stress data enabled the

calculation of a stability index, SQLCT. Altogether, eleven trials were performed during 2000/2001 and 2001/2002, testing several

weak layer types exhibiting a wide range of strengths. Weak layer strength and slab stress conditions varied widely across the

sampling sites, with coefficients of variation in weak layer shear strength ranging from 10% to 50%, coefficients of variation in

shear stress from 2% to 48%, and stability indices ranging from 1.8 to 5.7. Of the 54 snowpits completed, 10 pits were empirically

rejected as unrepresentative of the stability index at their sampling sites. Of the remaining 44 statistically analyzed pits, 33 pits

were statistically representative of their site-wide stability index, and the other 11 pits were found statistically unrepresentative of

their site. All five snowpits at a site were statistically representative of their site-wide stability index in three of the eleven trials.

The frequent inability of single pits to reliably represent stability on those eleven 900 m2 sampling sites, located on apparently

‘uniform’ slopes, highlights the importance of improving our understanding of the processes affecting the variability of snowpack

stability on any given day and the uncertainties associated with ‘point’ stability data.

D 2004 Published by Elsevier B.V.

Keywords: Avalanche forecasting; Stability; Extrapolation; Spatial variation

1. Introduction

Avalanche forecasting has been described as ‘‘. . .the prediction of current and future snow instability in

space and time relative to a given triggering (defor-

0165-232X/$ - see front matter D 2004 Published by Elsevier B.V.

doi:10.1016/j.coldregions.2003.12.003

* Corresponding author. Tel.: +1-970-387-5080; fax: +1-970-

387-5082.

E-mail address: [email protected] (C. Landry).

mation energy) level . . .’’ (McClung, 2002). It fol-

lows, then, that among the many objectives of a

forecaster is to ‘‘. . . minimize the uncertainty about

instability introduced by the temporal and spatial

variability of the snow cover (including terrain influ-

ences) . . .’’ (McClung, 2002).

Avalanche forecasters seek a variety of data in

order to minimize uncertainty regarding instability.

Evidence of instability obtained from the observation

C. Landry et al. / Cold Regions Science and Technology 39 (2004) 205–218206

of actual avalanches is considered relevant, ‘low

entropy’ (unambiguous) information collected at the

appropriate scale. Such data are most effective in

reducing uncertainty and, as such, are given the

highest weighting among the multiple and redundant

types of data collected (LaChapelle, 1980). In the

absence of actual avalanche observations, or to cor-

roborate the evidence they present, field measure-

ments of snowpack stability obtained from in-situ

stability tests are also considered relevant Class I data

(McClung and Schaerer, 1993). In-situ stability tests

measure a critical load required to cause snowpack

fracture in a small snowpack sample.

However, it is often unsafe to conduct in-situ

stability tests within avalanche starting zones, partic-

ularly when conditions approach the threshold of

avalanching (Fohn, 1987; CAA/NRCC, 1995). Fur-

ther, it is infeasible to obtain stability test data from

every starting zone of interest, given the magnitude of

terrain that most forecasters evaluate (Armstrong,

1991). For those reasons, avalanche forecasters rou-

tinely perform stability tests at carefully selected and

safe sites presumed to be ‘representative’ and predic-

tive of snowpack conditions in nearby avalanche

terrain, and then extrapolate their results (McClung

and Schaerer, 1993; Fredston and Fessler, 1994). For

instance, Chalmers and Jamieson (2001) correlate a

stability index collected at a level study plot to skier

triggered avalanching within 100 km. Skilled ava-

lanche forecasters are clearly capable of reliably

extrapolating sparse data over broad regions, probably

because of their ability to effectively filter large

amounts of diverse data (McClung, 2002).

The underlying premise of this stability-sampling

practice is that if and/or when a ‘representative’ slope

is a good proxy for snowpack and stability conditions

in nearby avalanche terrain, then reliable extrapolation

of stability test results is facilitated. According to a

fundamental tenet of geography, the closer the location

and characteristics of the stability test site are in space

to the location and characteristics of the extrapolated

point, the more successful extrapolation should be. It

follows, then, that extrapolating stability test results

from one location on an apparently ‘uniform’ slope

(selected to minimize known sources of spatial vari-

ability on snowpack properties across the slope) to a

nearby point on the same ‘uniform’ slope should be the

most reliable form of stability extrapolation possible.

While actual avalanches present comparatively un-

ambiguous stability information, stability test results

may contain substantial informational ambiguity re-

garding strength/stress relationships caused by the

unknown scale of spatial variations in snowpack

characteristics within the stability-sampling site itself.

Several studies have documented spatial variation in

snowpack stability within actual avalanche terrain,

wherein terrain and snowpack characteristics were

known to vary (Bradley, 1970; Conway and Abraham-

son, 1984; Fohn, 1989; Jamieson and Johnston, 1993a;

Birkeland, 2001; Kronholm et al., 2002; Stewart and

Jamieson, 2002).

Less attention has been given, however, to varia-

tion within study slopes specifically selected to max-

imize the chances of sampling a snowpack that is

homogeneous throughout the sampling site. Our study

investigated spatial variations in snowpack weak layer

strength and stability indices across sampling sites

specifically selected to minimize the effects of spatial

variations in terrain, aspect, substrate, vegetation and

wind on snowpack processes. We tested short-range

extrapolations of weak layer strength and stability

measurements across sampling sites in order to assess

whether a set of 10 (quantified loaded column)

stability tests from a single snowpit could reliably

predict the stability index of an apparently uniform

slope.

In addition to varying across space at a given

moment in time, stability also changes through time

at a given location in space. This study also measured

temporal variations in stability in a series of three 900-

m2 trials conducted at three side-by-side locations on

the same slope, over a period of 18 days. Those

results, and their interpretation, are the subject of a

companion article (Birkeland and Landry, 2002). The

current article will focus on the spatial variations in

snowpack strength and stability indices observed

during this study.

2. Methods

2.1. Stability sampling design and site selection

We adopted a systematic sampling design for this

study distributing five snowpits in a regular pattern

across a 30� 30 m, or 900 m2 test site (Fig. 1). Sys-

Fig. 1. Showing the 900 m2 stability-sampling site pit layout. The five sampling snowpits are numbered and represented as rectangles (see

Landry, 2002 for exact pit location coordinates and pit QLCT layout).

C. Landry et al. / Cold Regions Science and Technology 39 (2004) 205–218 207

tematic sampling assured coverage throughout the

site.

Seven 900 m2 sampling sites were selected based

on several desired attributes: no prior disturbance by

skiers, snowmobiles, etc.; planar slope profile; no or

minimal vegetation, besides grass; adequate distance

from nearby trees to prevent shading, tree-drop

‘bombs’ and other vegetation effects; smooth sub-

strate, without large scree or protruding bedrock;

slope angle from 25j to 30j, for safety; protection

from wind. Satisfying all of these conditions was

difficult, and some sites suffered from more wind

exposure than other, well-sheltered sites.

Nonetheless, in the opinion of the observers

conducting the trials, all of whom were experienced

avalanche forecasters, the selected sites seemed to

be nominally ‘uniform’ slopes capable of exhibiting

consistent snowpack characteristics throughout the

900 m2 stability-sampling site. Three of the stability

sampling sites were in the Bridger Range, northeast

of Bozeman, Montana, three in the Madison Range,

southwest of Bozeman, and the final site was at

Rogers Pass, British Columbia, near Fidelity Station

in Glacier National Park.

2.2. Measuring and calculating stability

Shear strength data for a single weak layer was

collected at each site using the quantified loaded

column stability test (QLCT) method (Landry et al.,

2001) (Fig. 2). Ten QLCT were performed in each of

the five snowpits at a site, using two rows of five, 50

cm-wide test cells, with the front of the second row of

five cells 1 m uphill of the front of the first row.

QLCT results were size-adjusted (Landry et al., 2001;

Jamieson and Johnston, 2001; Fohn, 1987) to calcu-

late shear strength sl. Slab properties above the weak

layer were also measured once at each pit in order to

calculate the shear stress sSlab acting upon the weak

layer at that pit location:

sSlab ¼ qghsinw ð1Þ

Fig. 2. A surface mode QLCT being performed during the 27 Jan.

2001 stability-sampling trials at the Bradley Meadow site in the

Bridger Range, MT. The observer is applying a rapid vertical load to

a 0.08 m2 isolated column of snow at the centroid of a plywood load

plate using a mechanical force gauge (Landry et al., 2001).


where h represents the thickness of the slab (m),

measured perpendicular to the slope, r is the density

of the slab (kg/m3), g is gravity, and w is slope angle.

A QLCT strength/stress stability index, similar in

principle but not strictly comparable to other stability

ratios and indices (e.g., Roch, 1966; Fohn, 1987;

Jamieson and Johnston, 1993b) due to differences

between the QLCT and shear frame methods, was

calculated:

SQLCT ¼ sl=sSlab ð2Þ

Mean and/or median stability ratio SQLCT values

were calculated for each pit and for each sampling site

(by pooling all valid SQLCT results).

2.3. Data analysis

Coefficients of variation for strength, stress, and

the stability index (of the general form CV= s/x,

where s represents the sample standard deviation

and x the sample mean) were calculated within

individual pits and across entire sites. Jamieson and

Johnston (2001) found that 20 of 28 sets of shear

frame measurements exhibited normal distributions of

shear strength, and recommended the coefficient of

variation as the best measure of variability in shear

strength since the standard deviation of shear strength

is known to increase as mean shear strength increases

(Jamieson and Johnston, 2001). Some of our trials

also exhibited normal distributions in the stability

index, when the results from all five pits at the site

were pooled (Birkeland et al., in press). However, the

small number of strength and stability index measure-

ments (nominally 10) at any individual pit did not

allow us to conclusively show that our pit-wise results

were normally distributed.

In order to compare data from pits to the pooled

data collected at a site, we used the non-parametric

Mann–Whitney test to evaluate the hypothesis of ‘‘no

difference’’ between pit-wise and site-wise shear

strength and stability index results. We pooled results

from a single snowpit with the remaining four snow-

pits at a site to obtain site stability SQLCT(Site) or site

shear strength sl(Site). If a particularly strong/weak or

stable/unstable pit were not pooled with the remaining

four pits, site-wide variability in strength or stability

would have been understated and made to appear

more consistent than was actually measured. Thus,

our analyses conservatively evaluated whether the

strength and/or stability in any single snowpit within

a site reliably represented site-wide strength and/or

stability and, therefore, whether that study site repre-

sented a single strength (or stability) population.

3. Results

3.1. Stability-sampling trials

Altogether, eleven 900 m2 sampling trials were

performed over the course of the 2000/2001 and 2001/

2002 winter seasons yielding data from 54 pits

(Table 1). During the Round Hill trial the entire slope

collapsed with a loud ‘whumpf’ during the prepara-

tion of the final pit (pit #5) and no data were obtained.

Hence, we completed 54 total pits rather than a full set

of 55.

In ten of the 11 trials, the weak layer tested was,

in fact, the weakest weak layer within the snow-

pack. In the eleventh, at Baldy Mountain on 18 Feb

2001, a total of five ‘weakest’ weak layers were

eventually identified during the course of the trial.

To provide a margin of safety for the sampling

teams, we attempted to avoid sampling sites known

to be approaching a state of instability susceptible

Table 1

Stability-sampling trials summary. Weak layer types are: ‘DH’= depth hoar; ‘BF’= basal facets; ‘NF’= new forms; ‘SH’= surface hoar;

‘NSF’= near-surface facets

Site (weak layer type) Trial date Site median

stability

index

CV site

stability

Site mean

strength (Pa)

CV site

shear

strength

CV site

shear

stress

Total

depth

(m)

Weak

layer age

(days)

Bacon Rind (DH) 4 Jan. 2001 1.85 31% 530 32% 4% 0.48 g60

Bradley Mdw. (NF) 27 Jan. 2001 5.14 22% 590 23% 32% 1.24 7

Round Hill (SH) 4 Feb. 2001 2.05 44% 830 50% 10% 1.82 7

Baldy Mtn. (DH/BF) 18 Feb. 2001 nr nr 1130a 26%a 28%a 1.11 g75

Saddle Peak (DH/BF) 18 Feb. 2001 1.90 26% 1490 25% 16% 1.08 g75

Bradley Mdw. (DH/BF) 18 Feb. 2001 nr nr 1660b 21%b 48%b 2.16 g75

Bradley Mdw. (NSF) 17 Mar. 2001 3.00 27% 430 27% 2% 1.15 g5

Middle Basin (DH/BF) 7 Dec. 2001 2.07 24% 700 23% 12% 0.89 g14

Lionhead Mtn. (SH) 9 Jan. 2002 2.50 10% 380 10% 4% 1.20 14



‘CV’ indicates coefficient of variation. Total snowpack depth (HS) is given.a Five different weak layers were revealed during the trial. Strength, stress and stability data above are for 19 valid QLCT results obtained

from the targeted depth-hoar weak layer only.b Results are for 20 valid QLCT results in the targeted depth-hoar weak layer; 30 tests exceeded the range of the QLCT equipment.


to the substantial disturbance these sampling ses-

sions produced. Weak layer types tested included

depth hoar and/or basal facets (five trials), near-

surface facets (one trial), surface hoar (four trials),

and precipitation particles (one trial) (Table 1).

Weak layers ranged in age from 5 to 75 days,

and snowpack depth ranged from 48 to 216 cm

(Table 1). In most trials, shear strength was more

variable, sometimes by an order of magnitude, than

shear stress. However, in three trials—Bradley

Meadow on 27 Jan. 2001, Baldy Mountain on 18

Feb. 2001, and Bradley Meadow on 18 Feb.

2001—shear stress was somewhat more variable

than shear strength (Table 1). During two trials—

Baldy Peak and Bradley Meadow on 18 Feb.

2001—inconsistent QLCT results from the targeted

weak layer resulted in all pits being found empir-

ically unrepresentative of their sites. Since those

sites had been carefully selected to minimize known

sources of spatial variability, and to optimize the

chance of finding uniform snowpack characteristics

throughout an apparently ‘uniform’ slope, we chose

to include those empirically rejected pits in our

analyses of their representativeness of site-wide

strength and stability; we also report our results

without including those two sites. Excluding those

pits from our analysis would have understated the

variability we observed during this research.

3.2. Variability in shear strength

Among the snowpits in our study for which a valid

coefficient of variation in shear strength could be

calculated CVs¯l, ranged from 6% to 37%, with a

mean of 17%. While the QLCT and shear frame

methods do apply different types of stress to a

measured weak layer, these values are similar to

results reported by Jamieson and Johnston (2001).

They had a mean coefficient of variation in shear

strength of 18%, and a range from 4% to 54%, in 114

sets of shear frame tests conducted on slopes of at

least 35j. The QLCT also is capable of detecting low

variability in shear strength; at Lionhead Mountain we

obtained 5.7%VCVs¯lV 6.2% in six separate snow-

pits over two sampling trials. Further, during two side-

by-side trials of the QLCT and shear frame conducted

in buried surface hoar at Rogers Pass, British Colum-

bia, in March 2000, coefficients of variation for the

QLCT and shear frame were, respectively, 15% versus

13%, and 11% versus 12% (Landry et al., 2001).

3.3. Pit-to-site differences in shear strength

Of the 54 total pits performed, 10 pits (19%) were

deemed unrepresentative of site-wide strength based

on conclusive empirical evidence. Such evidence

included QLCT results (or the lack thereof) which

Fig. 3. Variations from median strength (1500 Pa), Saddle Peak

trials of 18 Feb. 2001 (CVsl(Site) = 25%). White bars represent

QLCT results from the first (front) row of five tests in a pit and

black bars represent results from the second (back) row (N= 40).

Fig. 4. Variations from median strength (470 Pa), Bacon Rind trials

of 4 Jan. 2001 (CVsl(Site) = 32%). White bars represent QLCT

results from the first (front) row of five tests in a pit and black bars

represent results from the second (back) row (N = 50).


could not be interpreted (Bradley Meadows, 18 Feb.

2001), or the presence of multiple ‘weakest’ weak

layers (Baldy Mountain, 18 Feb. 2001). A statistically

significant difference (a = 0.05) was found between

pit-mean and site-wide strength at 14 (26%) of the

remaining 44 pits. Thus, 30 (56%) of the 54 pits were

statistically representative (percentages do not equal

100% due to rounding errors). Alternatively, if the

rejected pits were not considered the proportion of

representative pits increased to 68% (30 out of 44).

Only two trials (Bradley Meadow on 17 Mar. 2001,

and Middle Basin on 7 Dec. 2001) yielded full sets of

five pits that were statistically ‘representative’ of site-

wide strength. The Bradley Meadows 17 Mar. 2001

trials tested a layer of near-surface facets formed by

diurnal recrystallization (Birkeland, 1998) lying un-

derneath a thin frozen-rain crust. The Middle Basin

trial tested laminated layers of depth hoar and basal

facets. One other trial produced four statistically

‘representative’ pits, the Lionhead Mountain trial of

9 Jan. 2002. At the other end of the spectrum, at

Round Hill three of the four completed pits were

found statistically unrepresentative of the site-wide

shear strength sl(Site).

Charts of individual QLCT measurements of

strength revealed several interesting patterns of spatial

variation from site-wide strength (Figs. 3 and 4).

These charts present an oblique view of all five pits

at a site, with a key indicating relative pit locations

(refer also to Fig. 1). Strength for each valid QLCT is

shown as an individual ‘bar’ rising above or descend-

ing below the site-median strength ‘surface’.

Three general types of pit-wise variation from site-

wide shear strength were observed, and the Saddle

Peak trials exemplified all three (Fig. 3). Aweak layer

of 1–2 mm faceted grains and mixed forms overlying

a stronger layer of faceted forms was tested, with ‘Q2’

(average, mostly smooth) shear fractures (Johnson and

Birkeland, 2002) occurring at the interface between

the two faceted layers approximately 20 cm above-

ground. Total snowpack depth was 108 cm. Pit 3 at

Saddle Peak, the cluster at the center of the site,

contained the strongest and the weakest individual

QLCT results from the entire site. This typified many

pits in which the extremes of above- and below-

average site-wide shear strength were both present

in the same pit. Interestingly, such strong and weak

results were often obtained from adjoining tests, as

seen in the front row of Pit 3. Pits 2 and 4 (lower right

and upper right clusters) were typical of pits showing

a consistent departure, either above or below, site-

wide strength. And, finally, Pits 1 and 5 (lower and

upper left clusters) were characteristic of pits exhibit-

ing modest scatter about the site-median strength.

Saddle Peak Pits 1, 3 and 5 were all statistically

representative of site-wide shear strength.

A chart of the Bacon Rind trials graphically

depicts the variability of strength observed there

(Fig. 4). Consistent ‘Q1’ shear fractures (unusually

C. Landry et al. / Cold Regions Science a

clean, planar, smooth) occurred at the top of a layer

of basal depth hoar approximately 17 cm above-

ground; total snowpack depth was 48 cm. Pit 3, the

center cluster, contains mostly above-average strength

results.

Table 2

Mann–Whitney tests of pit-to-site stability (a= 0.05), shown by site and

Site Trial date Pits statistic

representati

of site stabi

Bacon Rind 4 Jan. 2001 Pit 1 ( p= 0

Pit 2 ( p= 0

Pit 4 ( p= 0

Pit 5 ( p= 0

Bradley Meadow 27 Jan. 2001 Pit 3 ( p= 0

Pit 4 ( p= 0

Round Hill 4 Feb. 2001 Pit 3 ( p= 0

Baldy Mountaina 18 Feb. 2001

Saddle Peak 18 Feb. 2001 Pit 1 ( p= 0

Pit 2 ( p= 0

Pit 3 ( p= 0

Bradley Meadowb 18 Feb. 2001

Bradley Meadow 17 Mar. 2001 Pit 1 ( p= 0

Pit 2 ( p= 0

Pit 3 ( p= 0

Pit 4 ( p= 0

Pit 5 ( p= 0

Middle Basinc 7 Dec. 2001 Pit 1 ( p= 0

Pit 3 ( p= 0

Pit 4 ( p= 0

Pit 5 ( p= 0

Lionhead Mountain 9 Jan. 2002 Pit 1 ( p= 0

Pit 2 ( p= 0

Pit 3 ( p= 0

Pit 4 ( p= 0

Pit 5 ( p= 0


Pit 2 ( p= 0

Pit 4 ( p= 0

Pit 5 ( p= 0


Pit 2 ( p= 0

Pit 3 ( p= 0

Pit 4 ( p= 0

Pit 5 ( p= 0

Totals 33

61%

a Five different weak layers were revealed during the trial.b Only 20 valid QLCT results in the targeted weak layer; 30 tests excc ‘Representativeness’ results reflect estimation of 9 QLCT tests resul

3.4. Pit-to-site differences in stability

When the variability of shear stress at a sampling

site was low (i.e., CV < 10%), the spatial patterns of

pit-to-site differences in stability closely resembled

nd Technology 39 (2004) 205–218 211

pit number

ally

ve

lity

Pits statistically

unrepresentative

of site stability

Pits empirically

unrepresentative

of site stability

.271) Pit 3 ( p= 0.023)

.153)

.606)

.445)

.219) Pit 1 ( p= 0.014)

.885) Pit 2 ( p= 0.001)

Pit 5 ( p= 0.041)

.125) Pit 1 ( p= 0.002)

Pit 2 ( p= 0.024)

Pit 4 ( p= 0.013)

Pits 1–5

.173) Pit 4 ( p= 0.001)

.161) Pit 5 ( p= 0.023)

.055)

Pits 1–5

.223)

.087)

.942)

.336)

.724)

.565) Pit 2 ( p= 0.039)

.147)

.953)

.921)

.427)

.112)

.858)

.565)

.122)

.294) Pit 3 ( p= 0.048)

.104)

.241)

.178)

.351)

.068)

.968)

.219)

.766)

11 10

20% 19%

eeded the range of the QLCT equipment.

ts where strength exceeded the range of the equipment.

Fig. 5. Variations from median stability index (2.1), Round Hill

trials of 4 Feb. 2001 (CVSQLCT(Site) = 44%). White bars represent

SQLCT results from the first (front) row of five tests in a pit and black

bars represent results from the second (back) row (N = 37).


pit-to-site differences in shear strength. But, where

shear stress showed larger variations across a given

site, patterns of pit-to-site differences in stability

changed, as compared to pit-to-site differences in

strength.

Of the 54 total pits sampled, 10 pits (19%)

exhibited conclusive empirical evidence (described

in 3.3 above) of ‘unrepresentativeness’ of site-wide

Fig. 6. Box-plot of stability indices, Round Hill trials of 4 Feb. 2001. Pits

pits in parentheses below, and site-wide (plot) stability is shown on the far

wide stability index. Outliers are greater than 1.5 times the interquartile ran

than three times the interquartile range away from the ends of the box.

stability. We found no statistically significant pit-to-

site differences (a = 0.05) in stability in 33 (61%) of

the 54 total pits, given that the 10 empirically rejected

pits are retained in the sample set of 54 pits (Table 2).

Alternatively, if the rejected pits were not considered

the proportion of representative pits increased to 75%

(33 out of 44). Three of the 11 trials produced full sets

of five pits exhibiting no statistically significant

difference in pit-to-site stability (Bradley Meadow

on 17 Mar. 2001, Middle Basin, and Lionhead Moun-

tain on 26 Jan. 2002).

A chart of stability indices for the Round Hill trials

presents an example of large variations in the stability

index between halves of the site, despite generally

consistent results within individual pits, and a trend

across the site (Fig. 5). This was the only site where

we observed such a pattern. This trial was also unique

in that two separate observer teams, both highly

trained and experienced, collected QLCT data. A

layer of 4–6 mm buried surface hoar located approx-

imately 125 cm above-ground yielded consistent ‘Q1’

shears. Total snowpack depth was 182 cm. During

preparation of Pit 5, the final pit performed, the entire

slope and test site collapsed thereby precluding valid

results from Pit 5. Hence, no cluster of results is

shown in the upper-left corner of the chart.

1–4 results are shown with Mann–Whitney p-values for individual

right. Only Pit 3 was statistically representative (a= 0.05) of the site-ge away from the ends of the box, and the extreme values are greater

Fig. 7. Variations from median stability index (3.0), Bradley

Meadow trials of 17 Mar. 2001 (CVSQLCT(Site) = 27%). White bars

represent SQLCT results from the first (front) row of five tests in a pit

and black bars represent results from the second (back) row (N = 46).


Both Pits 1 and 3 produced similar stability in-

dices while Pits 2 and 4 were also similar but sub-

stantially higher than those of Pits 1 and 3. Our

analysis found Pit 3 statistically representative of the

site-wide stability index (Table 2). Box-plots of

stability results at individual pits at Round Hill reveal

why only Pit 3 was statistically representative of site-

wide stability (Fig. 6) and the p-values also indicate

Fig. 8. Box-plot of stability indices, Bradley Meadow trials of 17 Mar. 2

individual pits in parentheses below, and site-wide (plot) stability is sho

(a= 0.05) of the site-wide stability index. Outliers are greater than 1.5 tim

extreme values are greater than three times the interquartile range away f

how unrepresentative Pits 1, 2 and 4 were (Table 2

and Fig. 6). At Round Hill, shear stress was generally

consistent (CVs¯Slab = 10%), but strength was clearly

not consistent throughout the site, with a standard

deviation of 420 Pa from a site-wide strength of 830

Pa (CVsl= 50%).

In contrast to Round Hill, a chart of stability indices

for the Bradley Meadow trials of 17 Mar. 2001 exhibits

more scatter about site-wide stability within the pits

(Fig. 7). Here, a layer of < 0.5 mm near-surface faceted

grains located 95 cm above-ground and immediately

below a 1–2 mm ice lens formed during a freezing rain

event yielded consistent ‘Q1’ shears. Total snowpack

depth was 115 cm. Shear stress was consistent

throughout the Bradley Meadow site on 17 Mar.

2001, varying by only 5 Pa (CVsSlab = 2%), but shear

strength was less consistent, with a standard deviation

of 120 Pa from a site-wide strength of 430 Pa (CVsl=

27%). Variations in stability indices at Bradley Mead-

ow on 17 Mar. 2001 (Fig. 7) therefore reflected

variations in strength.

Nevertheless, our analyses found all five pits at

Bradley Meadow on 17 Mar. 2001 statistically repre-

sentative of site-wide stability, showing that in-pit

variability did not preclude a pit from being represen-

tative of a site. Box-plots of the Bradley Meadow

001. Pits 1–5 results are shown with Mann–Whitney p-values for

wn on the far right. All five pits were statistically representative

es the interquartile range away from the ends of the box, and the

rom the ends of the box.


trials of 17 Mar. 2001 also reveal the variability in

stability both within and among pits (Fig. 8), yet their

p-values indicate the pits’ representativeness of site-

wide stability (Fig. 8 and Table 2).

4. Discussion

4.1. Potential sources of variability

In some trials, surprising variations in strength,

rather than in shear stress, resulted in poor represen-

tation of site-wide stability by individual pits. For

instance, our first trial at Bacon Rind involved an

extremely simple snowpack consisting of a 20-cm

thick layer of depth hoar covered by a 30-cm single-

layer slab, and initially consistent shear strength and

quality. Not until Pit 3, at the center of the site, did we

discover what seemed to be ‘anomalous’ variations in

strength, with the site’s strongest snow thus far (Figs.

4 and 9). No apparent spatial variations in the sub-

strate, vegetation, aspect, wind effects, or slope shape

was observed that might have explained that variabil-

ity at Bacon Rind at Pit 3.

Overall, we believe our site selection for these

eleven stability-sampling trials was successful in min-

Fig. 9. Box-plot of stability indices, Bacon Rind trials of 4 Jan. 2001. Pits

pits in parentheses below, and site-wide (plot) stability is shown on the fa

site-wide stability index. Outliers are greater than 1.5 times the interquarti

greater than three times the interquartile range away from the ends of the

imizing the influence of variations in terrain and

substrate on weak layer strength, at least to the extent

that we understand the sensitivity of the snowpack

over space and time to small differences in those

variables (Birkeland and Landry, 2002). Basal weak

layers, such as depth hoar, may be the most sensitive

to small differences in geothermal heating or in the

snowpack’s substrate, even when the substrate

appears uniform, while ‘higher’ weak layers forming

at the snowpack surface, farther from the ground, may

be less sensitive. However, our results at Round Hill,

on 4 Feb. 2001, provide evidence that weak layers

formed well above the ground surface can also exhibit

considerable spatial variation in strength; at this site

the targeted surface hoar layer was located approxi-

mately 125 cm above the ground.

Variations in the load on a weak layer produced by

variations in the overlying slab within a site appeared

to explain the variability in strength we observed

during other trials. Wind-drift effects were presumed

to cause those variations, particularly at the Bradley

Meadow and Baldy Mountain sites during the 18 Feb.

2001 trials. Chalmers and Jamieson (2001) found

evidence of increases in strength and stability in

surface hoar associated with long-term increases in

slab load, and Johnson and Jamieson (2000) made a

1–5 results are shown with Mann–Whitney p-values for individual

r right. Only Pit 3 was statistically unrepresentative (a= 0.05) of thele range away from the ends of the box, and the extreme values are

box.


similar finding for faceted weak layers. Although

those studies measured increases in strength and

stability associated with increasing loads over time,

the effect of spatial variations in the slab at a given

moment in time could help explain spatial variations

in weak layer strength and slope stability. However,

large spatial variations in weak layer strength also

occurred even when shear stress was effectively

uniform across a site. For instance, shear stress across

the Bacon Rind site varied only slightly, from 240 to

270 Pa, while site-wide shear strength ranged from

300 to 1140 Pa, and from 460 to 1140 Pa within a

single pit (Pit 3).

Variation in our observers’ QLCT technique, or

produced by the QLCT procedure itself, might have

offered yet another explanation for the variability in

strength we observed. We compared the variability

of our QLCT results to a study of variability in

shear frame test results and, even though the QLCT

method leaves all or a portion of the slab in place

above the weak layer, the coefficients of variation

in strength using the QLCT method closely resem-

ble those obtained by Jamieson and Johnston

(2001) using shear frames. This suggests that the

QLCT method may be no more prone to operator-

induced variations in test results than the shear

Fig. 10. Box-plot of stability indices, Saddle Peak trials of 18 Feb. 200

individual pits in parentheses below, and site-wide (plot) stability is shown

(a= 0.05) of the site-wide stability index. Outliers are greater than 1.5 tim

extreme values are greater than three times the interquartile range away f

frame method. Further, with six pits producing

coefficients of variation in shear strength of 6%

during the Lionhead trials of 9 Jan. 2002 and 15

Jan. 2002, at least some of which must be attrib-

uted to actual variations in snow strength, our

results show that the QLCT is capable of detecting

low levels of variability in comparatively weak

layers. Using the QLCT method, we were also able

to detect differences in shear strength at several

spatial scales: between side-by-side tests (Pit 3 at

Saddle Peak—Figs. 3 and 10), between pits (Pits 2

and 4 at Saddle Peak—Figs. 3 and 10), and within

a site (Round Hill—Fig. 5). Therefore, we conclude

that when performed by an expert, the QLCT

method, like the shear frame, can reliably measure

an index of shear strength without introducing

overly problematic levels of ‘background noise’

into the test results (Fig. 10).

Finally, we explored the relationship between the

age of a particular weak layer and its variability in

strength and stability. Perhaps, the older the weak

layer on our apparently uniform slopes had become,

the more opportunity they had to experience and

reflect spatially differing effects from variations in

the overlying slab and from subtle variations in the

underlying terrain and snowpack creep. The hypoth-

1. Pits 1–5 results are shown with Mann–Whitney p-values for

on the far right. Only Pits 1, 2 and 3 were statistically representative

es the interquartile range away from the ends of the box, and the

rom the ends of the box.


esis that a young surface hoar weak layer should be

less likely to exhibit variability in strength than

another older layer was belied by our results from

Round Hill, where buried surface hoar only 7 days old

produced CVsl(Site) = 50%, perhaps reflecting a rapid

response to variations in initial conditions in the slope

or weak layer and resulting in the spatial trend seen in

the results (Fig. 5). That result can be juxtaposed to the

three Lionhead Mountain trials in older buried surface

hoar, with values for CVsl(Site) of 10%, 10%, and

18% at ages 14, 20 and 31 days, which do lend support

to the hypothesis of increasing variability with in-

creased weak layer age. Our results in depth hoar/basal

facet weak layers do show a more consistent relation-

ship between weak layer age and variability of

strength. Since we have a limited amount of data for

each weak layer type, and our results show contradic-

tory patterns, we drew no conclusions regarding a

relationship relating weak-layer age to variability in

strength.

4.2. Stability tests as Class I data

Our findings may provide new insights regarding

the principle that Class I stability test results are more

easily interpreted than observations such as a snow-

pack profile (LaChapelle, 1980; McClung and Scha-

erer, 1993) and may suggest the need to re-visit how

stability evaluation is taught to the general public and

aspiring avalanche professionals. Experienced ava-

lanche forecasters selectively and conservatively con-

fer ‘reliability’ and ‘representativeness’ to their field

observations of stability. Further, Class II (snowpack

characteristics) information gleaned in the course of

conducting stability tests, such as patterns in the

snowpack stratigraphy (McCammon and Schweizer,

2002), or the mere presence/absence of a weak layer,

or the quality of shear fracturing (Johnson and Birke-

land, 2002), may also receive equal or greater weight-

ing than Class I stability test results (Schweizer and

Weisinger, 2001).

Since experience and expert knowledge are re-

quired to correctly interpret and appropriately apply

stability test results, novice backcountry travelers may

be inherently ill-equipped to interpret stability test

results. Nonetheless, the concept of the ‘representative

location’ for snowpits and stability tests is described

for, taught to, and commonly adopted by inexperi-

enced backcountry travelers, as well as aspiring ava-

lanche professionals and mountain guides (McClung

and Schaerer, 1993; Fredston and Fessler, 1994;

Tremper, 2001). Our results suggest the need for

increased emphasis on and awareness regarding the

conservative use of ‘representative’ sites and stability

test results.

4.3. Representative slopes and extrapolation

While professional avalanche forecasters generally

assume spatial variation in stability to be the norm in

complex terrain, our study shows that problematic

spatial variation in snowpack stability may also exist

on apparently nominal, ‘uniform’ slopes. When sta-

bility parameter measurements and stability indices

cannot be reliably extrapolated within an apparently

uniform and presumably ‘representative’ sampling

slope, uncertainty may not be reduced by extrapolat-

ing data obtained at the ‘representative slope’ to

surrounding terrain.

On the other hand, the fact that three of our trials

yielded five out of five pits representative of stability

(a = 0.05) provides evidence that single pits on care-

fully selected, apparently ‘uniform’ slopes, and/or in

time-tested study plots, sometimes do provide reliable

measures of snowpack characteristics throughout the

slope or plot. This result affirms the successful corre-

lation of study plot stability indices to regional skier-

triggered avalanching by Chalmers and Jamieson

(2001) and other studies of extrapolation (Fohn,

1987). Further, with the complications arising from

the dynamics of snow on slopes, more consistent

measurements may be possible in level areas than

on slopes, as Jamieson and Johnston (1993b) found

during their analyses of extrapolation of stability

parameters.

Our results lend support to the notion that sys-

tematic and/or random sampling of a presumably

representative study plot, or other apparently uniform

slope, in pursuit of ‘mean’ slope stability informa-

tion, may not always be as effective in reducing

uncertainty as seeking worst-case, ‘instability’ data

through ‘targeted sampling’ (McClung, 2002). The

ongoing challenge for avalanche forecasters is to

learn how to reliably predict those occasions when

a single sample will reliably represent an extensive

space and/or, alternatively, to learn how to objective-


ly interpret the results of a single pit, inferring or

deducing the spatial scale of variation in stability

extant on that day, both within the sampled slope

and beyond.

5. Conclusions

This research tested the hypothesis that a stability

index obtained from a set of 10 stability tests per-

formed at a single snowpit, which was located within

a carefully selected and apparently uniform site,

would reliably demonstrate a significant probability

of predicting the stability index of the entire site. After

11 stability-sampling trials at seven different sites a

conservative analysis or our results, retaining ten

empirically unrepresentative pits in the sample set,

found that 61% of our pits showed no statistically

significant difference (a = 0.05) between the pit sta-

bility index and the site-wide stability index whereas

the other 39% of our pits were either statistically or

empirically unrepresentative of their respective site’s

stability index. An alternative interpretation of our

results, in which ten empirically unrepresentative pits

were withdrawn from the sample set, showed that

75% of our pits were statistically representative

(a = 0.05) of their site-wide stability indices, while

25% were not.

Three of the eleven trials produced results in

which all five pits were representative of their site-

wide stability index. We did not detect so-called

deficit zones (Conway and Abrahamson, 1984). It

was not our purpose to establish a relationship

between our stability index SQLCT and actual ava-

lanche activity on our sampling days (Fohn, 1987).

Rather, we attempted to optimize the rigorous ex-

trapolation of a strength/stress stability index across

reasonably uniform slopes. Since our measurements

showed, at best, a one-in-four chance of misrepre-

senting a slope, we concluded that a single pit on an

apparently uniform slope was not shown to be a

statistically reliable predictor of sampling-site stabil-

ity parameters such as weak layer strength during

our trials.

Combining stability test results with other obser-

vations, such as snow stratigraphy and shear quality,

might reduce uncertainty. Irrespective of our results,

experienced avalanche forecasters can and do collect

stability data, and avalanche forecasts are largely

reliable and useful, in spite of the uncertainty of

stability test results (McClung, 2002). In part, this is

because human avalanche forecasters utilize a great

deal of sometimes redundant data to reduce their

uncertainty (LaChapelle, 1980).

Additional research is clearly needed to explain

why apparently uniform slopes sometimes do and

other times do not exhibit uniform stability that

sometimes can and other times cannot be reliably

sampled with a single set of stability tests. A

deeper understanding of the complex processes

leading to those conditions, and how they change

scales through time, would contribute to reducing

uncertainty regarding the spatial and temporal vari-

ability of snowpack stability (Birkeland and Landry,

2002).

Acknowledgements

We gratefully acknowledge support of this research

by the Montana State University Department of Earth

Sciences and the Barry C. Bishop Scholarship for

Mountain Research, National Science Foundation and

EPSCoR MONTS, American Avalanche Association,

Canadian Avalanche Association, Mazamas, the

American Alpine Club, and the Geological Society

of America Foundation’s John Montagne Fund.

Bridger Bowl Ski Area, the Parks Canada avalanche

control program at Glacier National Park, Life Link

International, and Snowmetrics contributed access to

research sites and equipment. Many thanks to our

Assistant Editor Jurg Schweizer, and to two anony-

mous reviewers. Their efforts led to substantial

improvements and refinements in the article. Frequent

discussions with Kalle Kronholm, Christine Pielmeier,

Pascal Hageli, and Jeff Deems also enhanced our

understanding of these results. Special thanks also go

to Jeff Deems and Ron Johnson for leading data

collection teams, and to Chuck Lindsay, Doug Chabot,

Stuart Dominick, Lance Riek, Chas Day, Aleph

Johnston-Bloom, Zack Matthews, Mark Schaffer,

Jim Rasmussen, John Kelly, Johann Schleiss, Dan

Miller, Michael Cooperstein, Jeanette Romig, Blase

Reardon, and the late Reid Sanders for their good work

and cheerful company despite always long days and

sometimes miserable conditions.


References

Armstrong, R.L., 1991. Snow properties research and data manage-

ment. Snow Science: Reflections on the past, Perspectives on

the Future. Proceedings of the Alta symposium, The Center for

Snow Science. Alta, Utah, pp. 1–9.

Birkeland, K.W., 1998. Terminology and predominant processes

associated with the formation of weak layers of near-surface

faceted crystals in the mountain snowpack. Arct. Alp. Res. 30

(2), 193–199.

Birkeland, K.W., 2001. Spatial patterns of snow stability throughout

a small mountain range. J. Glaciol. 47 (157), 176–186.

Birkeland, K.W., Landry, C.C., 2002. Changes in spatial patterns of

snow stability through time. Proceedings of the 2002 Interna-

tional Snow Science Workshop, Penticton, British Columbia,

Canada. British Columbia Ministry of Transportation, Ava-

lanche Program, Victoria, British Columbia, pp. 482–489.

Birkeland, K., Kronholm, K., Schneebeli, M., Pielmeier, C., in press.

Changes in the hardness and shear strength of a buried surface

hoar layer. Ann. Glaciol., 38.

Bradley, C., 1970. The location and timing of deep slab avalanches.

J. Glaciol. 9 (56), 253–261.

CAA/NRCC, 1995. Weather observation sites and procedures.

Appendix A, Observation Guidelines and Recording Stand-

ards for Weather, Snowpack and Avalanches. NRCC Techni-

cal Memorandum, vol. 132. Canadian Avalanche Association,

Revelstoke, British Columbia, pp. 61–62.

Chalmers, T.S., Jamieson, J.B., 2001. Extrapolating the skier sta-

bility of buried surface hoar layers from study plot measure-

ments. Cold Reg. Sci. Technol. 33 (2–3), 163–177.

Conway, H., Abrahamson, J., 1984. Snow stability index. J. Gla-

ciol. 30 (106), 321–327.

Fohn, P.M.B., 1987. The stability index and various triggering

mechanisms. Avalanche Formation, Movement and Effects.

IAHS Publ., vol. 162, pp. 195–214.

Fohn, P.M.B., 1989. Snowcover stability tests and the areal vari-

ability of snow strength. Proceedings of the 1988 International

Snow Science Workshop, Whistler, British Columbia, Canada.

Canadian Avalanche Association, Revelstoke, British Columbia,

pp. 262–273.

Fredston, J., Fessler, D., 1994. Snow Sense—a guide to evaluating

snow avalanche hazard. Alaska Mountain Safety Center, An-

chorage, Alaska, USA, p. 116.

Jamieson, J.B., Johnston, C.D., 1993a. Experience with rutsch-

blocks. Proceedings of the 1992 International Snow Science

Workshop, Breckenridge, Colorado, USA. Colorado Avalanche

Information Center, Boulder, Colorado, pp. 150–159.

Jamieson, J.B., Johnston, C.D., 1993b. Shear frame stability param-

eters for large-scale avalanche forecasting. Ann. Glaciol. 18,

268–273.

Jamieson, B., Johnston, C.D., 2001. Evaluation of the shear frame

test for weak snowpack layers. Ann. Glaciol. 32, 59–69.

Johnson, G., Jamieson, J.B., 2000. Strength changes of layers of

faceted snow crystals in the Columbia and Rocky Mountain

snowpack climates in southwestern Canada. Proceedings of

the 2000 International Snow Science Workshop, Big Sky, Mon-

tana, USA. American Avalanche Association, Pagosa Springs,

Colorado, pp. 86–93.

Johnson, G., Jamieson, J.B., 2000. Strength changes of layers of

faceted snow crystals in the Columbia and Rocky Mountain

snowpack climates in southwestern Canada. Proceedings of

the 2000 International Snow Science Workshop, Big Sky, Mon-

tana, USA, pp. 86–93.

Johnson, R.F., Birkeland, K.W., 2002. Integrating shear quality into

stability test results. Proceedings of the 2002 International Snow

Science Workshop, Penticton, British Columbia, Canada. British

Columbia Ministry of Transportation, Avalanche Program, Vic-

toria, British Columbia, pp. 508–513.

Johnson, R.F., Birkeland, K.W., 2002. Integrating shear quality

into stability test results. Proceedings of the 2002 International

Snow Science Workshop, Penticton, British Columbia, Canada,

pp. 508–513.

Kronholm, K., Schweizer, J., Schneebeli, M., 2002. Spatial variabil-

ity of snow stability on small slopes. Proceedings of the 2002

International Snow Science Workshop, Penticton, British Co-

lumbia, Canada. British Columbia Ministry of Transportation,

Avalanche Program, Victoria, British Columbia, pp. 549–554.

LaChapelle, E.R., 1980. The fundamental processes in conventional

avalanche forecasting. J. Glaciol. 26 (94), 75–84.

Landry, C.C., 2002. Spatial variation in snow stability on uniform

slopes: implications for extrapolation to surrounding terrain. MS

thesis, Department of Earth Sciences, Montana State University,

Bozeman, Montana, USA, p. 248.

Landry, C.C., Borkowski, J., Brown, R.L., 2001. Quantified loaded

column stability test: mechanics, procedure, sample-size selec-

tion, and trials. Cold Reg. Sci. Technol. 33 (2–3), 103–121.

McCammon, I., Schweizer, J., 2002. A field method for identifying

structural weaknesses in the snowpack. Proceedings of the 2002

International Snow Science Workshop, Penticton, British Co-

lumbia, Canada. British Columbia Ministry of Transportation,

Avalanche Program, Victoria, British Columbia, pp. 477–481.

McClung, D.M., 2002. The elements of applied avalanche forecast-

ing: Part I. The human issues. Nat. Hazards 25, 111–119.

McClung, D.M., Schaerer, P., 1993. The Avalanche Handbook. The

Mountaineers, Seattle, Washington, USA, p. 272.

Roch, A., 1966. Les variations de la resistance de la neige. Proceed-

ings of the International Symposium on Scientific Aspects of

Snow and Ice Avalanches, 1965. IAHS Publ., vol. 69, pp. 86–99.

Schweizer, J., Weisinger, T., 2001. Snow profile interpretation for

stability evaluation. Cold Reg. Sci. Technol. 33 (2–3), 179–188.

Stewart, K., Jamieson, B., 2002. Spatial variability of slab stability

in avalanche starting zones. Proceedings of the 2002 Interna-


Canada. British Columbia Ministry of Transportation, Ava-

lanche Program, Victoria, British Columbia, pp. 544–548.

Stewart, K., Jamieson, B., 2002. Spatial variability of slab stability

in avalanche starting zones. Proceedings of the 2002 Interna-


Canada, pp. 544–548.

Tremper, B., 2001. Staying Alive in Avalanche Terrain. The Moun-

taineers, Seattle, Washington, p. 284.

Documents

Variations in snow strength and stability on uniform slopes · Variations in snow strength and stability on uniform slopes C. Landrya,*, K. Birkelanda,b, K. Hansena, ... Seven slopes